KVM vm goes to 50% utilization. hangs. needs reset.

[rayk_sland]

I've got 2 debian lenny kvm vms configured. One of them periodically goes into a hung mode and has to be stopped and restarted. the other never does.

symptoms: proxmox control reports 50% CPU utilization. can do nothing. just click stop and then start

differences between vms
the problem vm has much more disk space alocated 500GB as opposed to 32GB on the "no problem" vm
the problem vm uses ext2fs on lvm as opposed to just ext2fs on the "no problem" vm.

 

[dietmar]

Any hint in syslog?

 

[rayk_sland]

nope. none whatsoever.

 

[rayk_sland]

It may be I can get some information off the vnc console. I'll post it at next crash.

 

[rayk_sland]

at the end of a lot of hex codes I have this.

"kernel panic not syncing fatal exception in interrupt"

 

[rayk_sland]

dmesg has a complaint in it about memory -- need HIGHMEM kernel
---

it's 32 bit on 64 bit.

It's also running on virtio... is scsi/ide more stable? I imagine I'll have to change fstab...

 

[dietmar]

It's also running on virtio... is scsi/ide more stable? I imagine I'll have to change fstab...

Try ide (i consider that most stable).

 

[rayk_sland]

So. Just a matter of changing /boot/grub/menu.lst (possibly) and /etc/fstab first, and then changing the qemu config files and restarting the vm second? Also, I was thinking of redoing the vm as a 64bit debian, because of the warnings about HIGHMEM, and some hints on some changelog page (can't find it now) about stability problems with kvm, virtio and 32bitness. Do you have any comment on that?

Thanks for your help so far.

 

[rayk_sland]

Ok. I don't change fstab. Having selected ide as the main driver. I have no HIGHMEM complaints in dmesg but the main hard disk device shows up as /dev/vda. What's with that. I thought that naming convention comes with virtio!

 

[rayk_sland]

more.
device name did change. after a full qemu reload
what I did before was a shutdown -r inside the vm.
and now dmesg does report

[ 0.000000] Warning only 4GB will be used.
[ 0.000000] Use a HIGHMEM64G enabled kernel.
[ 0.000000] 3200MB HIGHMEM available.
[ 0.000000] 896MB LOWMEM available.

 

[rayk_sland]

Well we were stable for 9 days. And now. Same symptom. 50 utilization and boom! hung server. Is there a way to get at the console from a ssh session? Java doesn't always initialize to run vnc.

 

[dietmar]

virtio or ide?

 

[rayk_sland]

ide

 

[dietmar]

Please can you post your VM config (/etc/qemu-server/xxx.conf)?

 

[rayk_sland]

As requested

name: MailServer-2
ide2: debian-501-i386-netinst.iso,media=cdrom
smp: 2
vlan0: virtio=AE:F2:9A:4B:97:5B
bootdisk: ide0
ide0: vm-102-disk.qcow2
ostype: l26
memory: 4096
vlan1: virtio=CA3:0E:CD:A1:7C
onboot: 0

 

[dietmar]

You still use virtio - please try without.

 

[rayk_sland]

You still use virtio - please try without.

is virtio lan out too?

I thought the choice was between virtio and ide. Which obviously leaves lan out of the picture...

Which LAN driver should I use?

 

[gkovacs]

I also have a KVM Windows guest that goes to 50% utilization on a periodical basis, needs reset as well.

This has not happened before, only since KVM-86.

 

[dietmar]

Which LAN driver should I use?

try e1000

 

[rayk_sland]

name: MailServer-2
ide2: debian-501-i386-netinst.iso,media=cdrom
smp: 2
vlan0: e1000=AE:F2:9A:4B:97:5B
bootdisk: ide0
ide0: vm-102-disk.qcow2
ostype: l26
memory: 4096
vlan1: e1000=CA3:0E:CD:A1:7C
onboot: 0

Requested changes done. Very little difference. Server went 50% this morning. Slight change. VNC didn't display a kernel crash, just lots of errors about a hung process (exim) and I was almost able to log in, except that I didn't get a bash prompt and eventually had to stop the vm from the web gui anyways.

I've configured up a 64 bit vm as close as possible to the same config as this one (except lvm) and I'm monitoring it.